Neural architecture search (NAS) has achieved great success in different computer vision tasks such as object detection and image recognition. Moreover, deep learning models have millions or billions of parameters and applying NAS methods when considering a small amount of data is not trivial. Unlike computer vision tasks, labeling time series data for supervised learning is a laborious and expensive task that often requires expertise. Therefore, this paper proposes a simple-yet-effective fine-tuning method based on repeated k-fold cross-validation in order to train deep residual networks using only a small amount of time series data. The main idea is that each model fitted during cross-validation will transfer its weights to the subsequent folds over the rounds. We conducted extensive experiments on 85 instances from the UCR archive for Time Series Classification (TSC) to investigate the performance of the proposed approach. The experimental results reveal that our proposed model called NAS-T reaches new state-of-the-art TSC accuracy, by designing a single classifier that is able to beat HIVE-COTE: an ensemble of 37 individual classifiers.
Metaheuristic algorithms (MAs) have seen unprecedented growth thanks to their successful applications in fields including engineering and health sciences. In this work, we investigate the use of a deep learning (DL) model as an alternative tool to do so. The proposed method, called MaNet, is motivated by the fact that most of the DL models often need to solve massive nasty optimization problems consisting of millions of parameters. Feature selection is the main adopted concepts in MaNet that helps the algorithm to skip irrelevant or partially relevant evolutionary information and uses those which contribute most to the overall performance. The introduced model is applied on several unimodal and multimodal continuous problems. The experiments indicate that MaNet is able to yield competitive results compared to one of the best hand-designed algorithms for the aforementioned problems, in terms of the solution accuracy and scalability.
Metaheuristic algorithms have successfully tackled many difficult and ill-conditioned optimization problems. Nevertheless, performance of these methods is subjected to the complexity and fitness landscape of the problem at hand. Accordingly, designing metaheuristic algorithms that work well on a variety of optimization problems is not a trivial task. In this study, we introduce a novel framework for improving generalization capability of the metaheuristic algorithms based on the notion of gene expression programming (GEP). The proposed framework introduces a modified GEP (MGEP) in order to adaptively design search operators of a metaheuristic algorithm. During evolution process, a multi-criteria procedure determines the search operators that are preferable and can obtain high accuracy results. Performance of the proposed approach is empirically evaluated on CEC 2013 test suite. The obtained results confirm that the evolved metaheuristic algorithms by this framework perform similarly to or better than the standard versions.
This study focuses on the problem of optimum selection of tuned mass dampers (TMDs) parameters for buildings under seismic excitations. Vibration control of structure under seismic excitations for reduction of structural damages and comfort/serviceability considerations is addressed by employing a design process based on multi-objective cuckoo search (MOCS) for simultaneous reduction of seismic responses of the structures in the terms of displacement and acceleration of stories and tuning of TMD with a smaller mass ratio. The numerical studies are carried out on two benchmark buildings. In comparison with several other documented methods, the performance of the MOCS for the optimal design of TMDs is evaluated. The results show that the MOCS performs better than other strategies in term of the simultaneous reduction of maximum displacements and accelerations of the structures subjected to different seismic excitations. Furthermore, the feasibility of MOCS to choose a smaller mass, stiffness, and damping is assessed. The simulation results confirm that the MOCS is able to present a solution leading to a more practical and economic selection of TMD parameters with a choice of a smaller mass, stiffness, and damping so that it is capable of maintaining the desired level of the reduction of structural responses in different earthquake excitations.
The majority of evolutionary algorithms (EAs) adopt Pseudo-Random Numbers Generator (PRNG) to initialize their population. This can affect the behavior of an EA for high dimensional problems due to the curse of dimensionality and has been known as a potentially serious challenge. Therefore, intelligent initialization of individual candidates has been more explored recently. As a different approach, this study proposes a machine-learning based algorithm to address the aforementioned problem. The introduced SVM based Smart Sampling, we call as SVM-SS, employs Support Vector Machine (SVM) to discover promising regions faster. The proposed method and Differential Evolution (DE) are then combined to evaluate our approach. Numerical results on a set of classic benchmark functions show that the proposed algorithm performs better in comparison with several state-of-the-art population initialization methods. To examine the scalability of the SVM-SS, it is also applied on large scale optimization problems and such results were also in consonance with the previous experiments.
In this study, we provide mathematical and practice-driven justification for using $[0,1]$ normalization of inconsistency indicators in pairwise comparisons. The need for normalization, as well as problems with the lack of normalization, are presented. A new type of paradox of infinity is described.
'%3e%3cg id='e'%3e%3cg id='c' fill='none' stroke='%23fff' stroke-width='2'%3e%3cpath id='b' d='M0 1h26M1 10V5h8v4h8V5h-5M4 9h2m20 0h-5V5h8m0-5v9h8V0m-4 0v9' transform='scale(1.4)'/%3e%3cpath id='a' d='M0 7h9m1 0h9' transform='scale(2.8)'/%3e%3cuse xlink:href='%23a' y='120'/%3e%3cuse xlink:href='%23b' y='145.2'/%3e%3c/g%3e%3cg id='d'%3e%3cuse xlink:href='%23c' x='56'/%3e%3cuse xlink:href='%23c' x='112'/%3e%3cuse xlink:href='%23c' x='168'/%3e%3c/g%3e%3c/g%3e%3cuse xlink:href='%23d' x='168'/%3e%3cuse xlink:href='%23e' x='392'/%3e%3c/g%3e%3cg fill='%23da0000' transform='matrix(45 0 0 45 315 180)'%3e%3cg id='f'%3e%3cpath d='M-.548.836A.912.912 0 0 0 .329-.722 1 1 0 0 1-.548.836'/%3e%3cpath d='M.618.661A.764.764 0 0 0 .422-.74 1 1 0 0 1 .618.661M0 1l-.05-1L0-.787a.31.31 0 0 0 .118.099V-.1l-.04.993zM-.02-.85 0-.831a.144.144 0 0 0 .252-.137A.136.136 0 0 1 0-.925'/%3e%3c/g%3e%3cuse xlink:href='%23f' transform='scale(-1 1)'/%3e%3c/g%3e%3c/svg%3e)
'/%3e%3c/defs%3e%3cpath fill='%23012169' d='M0 0h10080v5040H0z'/%3e%3cpath stroke='%23fff' stroke-width='.6' d='m0 0 6 3m0-3L0 3' clip-path='url(%23b)' transform='scale(840)'/%3e%3cpath stroke='%23e4002b' stroke-width='.4' d='m0 0 6 3m0-3L0 3' clip-path='url(%23c)' transform='scale(840)'/%3e%3cpath stroke='%23fff' stroke-width='840' d='M2520 0v2520M0 1260h5040'/%3e%3cpath stroke='%23e4002b' stroke-width='504' d='M2520 0v2520M0 1260h5040'/%3e%3cg fill='%23fff'%3e%3cuse xlink:href='%23d' x='2520' y='3780'/%3e%3cuse xlink:href='%23a' x='7560' y='4200'/%3e%3cuse xlink:href='%23a' x='6300' y='2205'/%3e%3cuse xlink:href='%23a' x='7560' y='840'/%3e%3cuse xlink:href='%23a' x='8680' y='1869'/%3e%3cuse xlink:href='%23e' x='8064' y='2730'/%3e%3c/g%3e%3c/svg%3e)
